Download GANTMACHER-KRE˘IN THEOREM FOR 2 NONNEGATIVE OPERATORS IN SPACES OF FUNCTIONS

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Basis (linear algebra) wikipedia , lookup

Linear algebra wikipedia , lookup

Fundamental theorem of algebra wikipedia , lookup

Eigenvalues and eigenvectors wikipedia , lookup

Perron–Frobenius theorem wikipedia , lookup

Jordan normal form wikipedia , lookup

Invariant convex cone wikipedia , lookup

Dual space wikipedia , lookup

Hilbert space wikipedia , lookup

Symmetry in quantum mechanics wikipedia , lookup

Bra–ket notation wikipedia , lookup

Tensor operator wikipedia , lookup

Transcript
GANTMACHER-KREĬN THEOREM FOR 2 NONNEGATIVE
OPERATORS IN SPACES OF FUNCTIONS
O. Y. KUSHEL AND P. P. ZABREIKO
Received 26 June 2005; Accepted 1 July 2005
The existence of the second (according to the module) eigenvalue λ2 of a completely continuous nonnegative operator A is proved under the conditions that A acts in the space
L p (Ω) or C(Ω) and its exterior square A ∧ A is also nonnegative. For the case when the
operators A and A ∧ A are indecomposable, the simplicity of the first and second eigenvalues is proved, and the interrelation between the indices of imprimitivity of A and A ∧ A
is examined. For the case when A and A ∧ A are primitive, the difference (according to
the module) of λ1 and λ2 from each other and from other eigenvalues is proved.
Copyright © 2006 O. Y. Kushel and P. P. Zabreiko. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use,
distribution, and reproduction in any medium, provided the original work is properly
cited.
1. Introduction
In the monograph [3] the following statement was proved: if the matrix A of a linear
operator A in the space Rn is primitive along with its associated A( j) (1 < j ≤ k) up to the
order k, then the operator A has k positive simple eigenvalues 0 < λk < · · · < λ2 < λ1 , with
a positive eigenvector e1 corresponding to the maximal eigenvalue λ1 , and an eigenvector e j ,
which has exactly j − 1 changes of sign, corresponding to jth eigenvalue λ j (see [3, page
310, Theorem 9]). Matrices with mentioned features are called henceforth k-completely
nonnegative; in the most important case k = n they are called oscillatory.
Naturally, there arises a problem whether it is possible to extend this statement to
operators in infinite-dimensional spaces, for example, to linear integral operators. This
problem practically has not been studied in full volume. However, in the monograph [3],
Gantmacher and Kreı̆n have thoroughly studied the linear integral operators
Kx(t) =
Hindawi Publishing Corporation
Abstract and Applied Analysis
Volume 2006, Article ID 48132, Pages 1–15
DOI 10.1155/AAA/2006/48132
b
a
k(t,s)x(s)ds
(1.1)
2
Gantmacher-Kreı̆n theorem for 2 nonnegative operators
acting in the space L2 ([a,b]) with continuous kernels k(t,s), for which the matrices
k(ti ,t j )n1 (n = 1,2,...) for any points t1 ,...,tn ∈ [a,b], among which at least one is interior, are oscillatory. Such kernels, named in [3] oscillatory, form quite full analogue to
oscillatory matrices. In [3], for the integral operators with continuous oscillatory kernels,
it was proved that there exists a converging-to-zero sequence of positive simple eigenvalues λ1 > λ2 > · · · > λn > · · · with eigenfunctions en (t) that has exactly n − 1 changes of
sign, corresponding to the nth eigenvalue λn (see [3, page 211]).
In connection with the formulated Gantmacher-Kreı̆n theorem, there arises a natural
question on the possibility of spreading the statements about k-completely-nonnegative
matrices from [3] onto the integral operators with k-completely-nonnegative kernels,
that is, the kernels k(t,s), for which the matrices k(ti ,t j )n1 (n = 1,2,...,k) for any points
t1 ,...,tn ∈ [a,b], among which at least one is interior, are oscillatory. The answer to this
question is positive. Moreover, this statement was actually proved exactly in [3].
However, here arises a question how substantial the condition of continuity of the
kernel k(t,s) is in these statements and how substantial the assumption that the problem
is regarded in the space of functions, defined exactly on the interval [a,b], is. And of
course the natural question arises whether it is possible to obtain similar statements for
abstract (not necessarily integral) operators in an arbitrary Banach spaces.
In the present paper we study 2-completely-nonnegative (or otherwise bi-nonnegative) operators in the spaces L p (Ω) (1 ≤ p ≤ ∞) and C(Ω). As the authors believe, the
natural machinery for the examination of such operators is a crossway from studying an
operator A in one of the spaces L p (Ω) and C(Ω) to the study of the operators A ⊗ A and
A ∧ A, acting, respectively, in the spaces L p ⊗ L p = L p (Ω × Ω) and L p ∧ L p = Lap (Ω × Ω)
(the latter is a subspace of the space L p ⊗ L p = L p (Ω × Ω), consisting of antisymmetric
functions, i.e., functions x(t,s), for which x(t,s) = −x(s,t)).
2. Tensor and exterior square of the spaces L p (Ω) and C(Ω)
Let (Ω, A,μ) be a triple consisting of some set Ω, some σ-algebra A of “measurable” subsets and some σ-finite and σ-additive measure on A. We will be interested in the space
L p (Ω) of functions, integrable on Ω with the power p for 1 ≤ p < ∞ or measurable and
substantially bounded for p = ∞, the analogous space L p (Ω × Ω) of functions, integrable
on Ω × Ω with the power p for 1 ≤ p < ∞ or essentially bounded for p = ∞ and, finally,
the subspace Lap (Ω × Ω) of the space L p (Ω × Ω) of antisymmetric functions. Henceforth
let p be a fixed number from [1, ∞].
We start with observing the following facts:
(1) the space Lap (Ω × Ω) is one of the tensor products of the space L p (Ω) by itself , and,
respectively,
(2) the space Lap (Ω × Ω) is one of the exterior products of the space L p (Ω) by itself.
The first of these statements means the following.
(a) For arbitrary functions x1 ,x2 ∈ L p (Ω) their -product x1 x2 (t1 ,t2 ) = x1 (t1 )x2 (t2 )
belongs to the space L p (Ω × Ω), with
x1 t1 x2 t2 = x1 t1 x2 t2 .
(2.1)
O. Y. Kushel and P. P. Zabreiko 3
(b) The linear hull of the set of all -products of functions from L p (Ω), that is, the
set of all functions of the form
x t1 ,t2 =
x1i t1 x2i t2
(2.2)
i
is dense in the space L p (Ω × Ω).
The second statement means the following.
(a) The ∧-product of arbitrary functions x1 ,x2 ∈ L p (Ω) with x1 ∧ x2 (t1 ,t2 ) =
x1 (t1 )x2 (t2 ) − x1 (t2 )x2 (t1 ) also belongs to the space L p (Ω × Ω), and it is obvious
that
x1 ∧ x2 t1 ,t2 = −x1 ∧ x2 t2 ,t1 ,
x1 ∧ x2 t1 ,t2 ≤ 2x1 t1 x2 t2 .
(2.3)
(b) The linear hull of the set of all ∧-products of the functions from L p (Ω) is dense
in the space Lap (Ω × Ω).
The space Lap (Ω × Ω) is isomorphic in the category of Banach spaces to the space L p (W),
and (Ω × Ω) \ (W ∪ W)
have
where W is the subset Ω × Ω, for which the sets W ∩ W
= {(t2 ,t1 ) : (t1 ,t2 ) ∈ W } (such sets do always exist). Really, extendzero measure; here W
ing the functions from L p (W) as antisymmetric functions from W to Ω × Ω, we obtain
the set of all the functions from Lap (Ω × Ω). Further, setting the norm of a function in
L p (W) to be equal to the norm of its extension, we get that the spaces Lap (Ω × Ω) and
L p (W) are isomorphic in the category of normed spaces.
The general scheme of the interrelations between the spaces L p (Ω) ⊕ L p (Ω), L p (Ω ×
Ω), Lap (Ω × Ω), and L p (W) can be represented by the diagram
⊗
a
L p (Ω) ⊗ L p (Ω) −−→ L p (Ω × Ω) −−→ Lap (Ω × Ω) = L p (W),
(2.4)
where a is the antisymmetrization operator acting from L p (Ω × Ω) to Lap (Ω × Ω) according to the rule
ax t1 ,t2 =
x t1 ,t2 − x t2 ,t1
.
2
(2.5)
Let us examine some examples of constructing the set W for different sets Ω.
(1) Let Ω = [a,b]; then Ω × Ω = [a,b]2 , and as W we may regard the triangle, de is defined by the infined by inequalities a ≤ t1 ≤ t2 ≤ b. Really, in this case W
and W ∩ W
= w0 , where
equalities a ≤ t2 ≤ t1 ≤ b, Ω × Ω = [a,b]2 = W ∪ W
w0 , defined by the inequalities a ≤ t1 = t2 ≤ b, is a set of zero measure.
(2) Consider another example. Let Ω = [a,b]2 . Then Ω × Ω = [a,b]4 . Define on
the space R2 the following order relation: (t1 ,t2 ) ≤ (s1 ,s2 ), if t1 ≤ s1 . Introduce
= {(t1 ,t2 ,t3 ,t4 ) ∈ [a,b]4 :
W = {(t1 ,t2 ,t3 ,t4 ) ∈ [a,b]4 : (t1 ,t2 ) < (t3 ,t4 )} and W
∪ w0 , where w0 = {(t1 ,t2 ,t3 ,t4 ) ∈
(t3 ,t4 ) < (t1 ,t2 )}. As we see, Ω × Ω = W ∪ W
[a,b]4 : (t1 ,t2 ) = (t3 ,t4 )} is a set of zero measure in the 4-dimensional space.
After thorough examination of the inequalities defining the set W, one obtains
4
Gantmacher-Kreı̆n theorem for 2 nonnegative operators
W = {a ≤ t1 < t3 ≤ b; a ≤ t2 , t4 ≤ b}. The geometrical interpretation of this construction is as follows: the 4-dimensional cube is divided by the hyperflat t1 = t3
into two symmetric parts. Notice that the cube can be divided in such a way with
the help of another surfaces as well, for example, with the help of the hyperflat
t1 + t2 = t3 + t4 .
(3) Suppose that on the set Ω a relation with the following properties is given:
(a) almost all the elements of Ω are comparable;
(b) μ({t1 ,t2 ∈ Ω : t1 t2 } ∩ {t1 ,t2 ∈ Ω : t2 t1 }) = 0. Then we can define the sets
= {(t1 ,t2 ) ∈ Ω2 : t2 t1 }, possessing the
W = {(t1 ,t2 ) ∈ Ω2 : t1 t2 } and W
necessary properties.
Now let (Ω,τ) be some compact Hausdorf topological space and let C(Ω) be the space
of all continuous functions on Ω. Then the set Ω × Ω, with the topology τ × τ given on
it, is also a compact Hausdorf space. Denote by C(Ω × Ω) the space of all continuous
functions on Ω × Ω, and by C a (Ω × Ω)—the subspace C(Ω × Ω) of all antisymmetric
functions on Ω × Ω. Just as in the case of Lebesgue spaces, the space C(Ω × Ω) is one
of the tensor products of the space C(Ω) by itself, and the space C a (Ω × Ω) is one of
the exterior products of the space C(Ω) by itself. In other words, for them the analogues
of statements (1) and (2) for Lebesgue spaces are true. Further, the space C a (Ω × Ω) is
= Δ, Δ =
isomorphic to the space C0 (W) (here W is a subset Ω × Ω, for which W ∩ W
{(t,t) : t ∈ Ω}, W ∪ W = Ω × Ω, and C0 (W) is a subspace of the space C(W), consisting
of all functions x(t1 ,t2 ), for which x(t1 ,t1 ) = 0). In particular, the following diagram is
true:
⊗
a
C(Ω) ⊗ C(Ω) −−→ C(Ω × Ω) −−→ C a (Ω × Ω) = C0 (W) ⊂ C(W).
(2.6)
Sometimes Lebesgue spaces and the space of all continuous functions have to be examined at the same time. In this case it is natural to require continuous functions to be
measurable. This means that the topology τ, σ-algebra A, and the measure μ on Ω must
be related to each other in the following way: A contains all the closed sets from τ and the
measure μ is regular, that is, for any A ∈ A and any number > 0 there exist a closed set
F and an open set G such that F ⊂ A ⊂ G and μ(G \ F) < . In this case the space C(Ω) is
associated with a closed subspace of the space L∞ (Ω).
3. Tensor and exterior squares of linear operators in the spaces L p (Ω) and C(Ω)
Let A and B be continuous linear operators acting in the space L p (Ω). These operators
generate the operator A ⊗ B in the space L p (Ω × Ω) as follows: on degenerate functions it
is defined by the equality
(A ⊗ B)x t1 ,t2 =
j
j j Ax1 t1 · Bx2 t2
x t1 ,t2 =
j j x1 t1 · x2 t2
,
(3.1)
j
and on arbitrary functions it is defined by extension via continuity from the subspace of
degenerate functions onto the whole of L p (Ω × Ω). The possibility of such an extension
is due to the density of the set of all degenerate functions in the space L p (Ω × Ω) and to
O. Y. Kushel and P. P. Zabreiko 5
the fact that the operator A ⊗ B is bounded on the subspace of degenerate functions of
the space L p (Ω × Ω); the latter comes out from the following observations.
Let
x t1 ,t2 =
j
j x1 t1 x2 t2
(3.2)
j
be a degenerate function from L p (Ω × Ω). It is obvious that for almost every t2 ∈ Ω this
function is measurable as a function of t1 ∈ Ω. Moreover, it can be regarded as a linear
j
j
combination of functions x1 (t1 ) with coefficients x2 (t2 ). Therefore
A(1) x t1 ,t2 =
j
j
Ax1 t1 · x2 t2 ;
(3.3)
j
the obtained function turns out to be measurable with respect to all the variables (the
operator’s A index (1) means that it is used for the variable t1 ). Because of the Fubini
theorem, the following estimate is true:
j
j
A(1) x t1 ,t2 ≤ A x t1 · x t2 )
.
1
2
(3.4)
j
j
j
Further, the function j Ax1 (t1 )x2 (t2 ) is measurable with respect to all the variables, it
is measurable as a function of t2 ∈ Ω for almost every t1 ∈ Ω, and it can also be regarded
j
j
as a linear combination of functions x2 (t2 ) with coefficients Ax1 (t1 ). Therefore, after using
the operator B for the variable t2 ,
B(2) A(1) x t1 ,t2 =
j
j
Ax1 t1 · Bx2 t2 ;
(3.5)
j
the obtained function turns out to be measurable with respect to all the variables. Applying the Fubini theorem again, we get the estimate
j
j B(2) A(1) x t1 ,t2 ≤ A B x1 t1 · x2 t2 .
(3.6)
j
We may conventionally write down the value of the operator A ⊗ B in the form of
(A ⊗ B)x t1 ,t2 = A(1) B(2) x t1 ,t2 = B(2) A(1) x t1 ,t2 ,
(3.7)
for arbitrary functions from L p (Ω × Ω) as well. However, with such a separate usage of
the operators A and B, there arises a question of measurability of the function B(2) x(t1 ,t2 )
for the variable t1 . To avoid this trouble, we have to use the procedure of extension by continuity from the subspace of degenerate functions, where, as shown above, the mentioned
trouble does not arise.
In the case of the space C(Ω × Ω), formula (3.7) and the estimate
(A ⊗ B)x t1 ,t2 ≤ A B x t1 ,t2 (3.8)
6
Gantmacher-Kreı̆n theorem for 2 nonnegative operators
arising from it are obvious, since the function that is continuous with respect to all the
variables is continuous also in each variable separately with the fixed another one.
Further in this work we will examine exclusively the tensor square A ⊗ A of the operator A.
Let us examine the operator A ∧ A : Lap (Ω × Ω) → Lap (Ω × Ω), defined as the restriction
of the operator A ⊗ A onto the subspace Lap (Ω × Ω). It is obvious that for degenerate
antisymmetric functions the operator A ∧ A can be defined by the equality
(A ∧ A)x t1 ,t2 =
j
j
Ax1 t1 ∧ Ax2 t2 ,
x t1 ,t2 =
j
j
j
x1 t1 ∧ x2 t2 .
(3.9)
j
Decompose L p (Ω × Ω) into the direct sum of subspaces invariant with respect to A ⊗ A:
L p (Ω × Ω) = Lap (Ω × Ω) ⊕ Lsp (Ω × Ω),
(3.10)
where Lsp (Ω × Ω) is the subspace of all symmetric functions from L p (Ω × Ω). The operator A ⊗ A can be represented in the block form
A∧A
0
.
A⊗A =
0
(A ⊗ A)|Lsp
(3.11)
Further, it will be useful to compare the operator A with the antisymmetrization of
its tensor square a ◦ (A ⊗ A) : L p (Ω × Ω) → Lap (Ω × Ω) ⊂ L p (Ω × Ω), where a is the antisymmetrization operator, defined by formula (2.5). Taking into account that the antisymmetrization operator leaves antisymmetric functions without changes, we conclude
that the restriction of a ◦ (A ⊗ A) onto the subspace Lap (Ω × Ω) coincides with A ∧ A.
The space C(Ω × Ω) can also be decomposed into the direct sum of subspaces invariant with respect to A ⊗ A:
C(Ω × Ω) = C a (Ω × Ω) ⊕ C s (Ω × Ω),
(3.12)
where C s (Ω × Ω) is the subspace of all symmetric functions from C(Ω × Ω). It is easy
to see that the operator A ⊗ A : C(Ω × Ω) → C(Ω × Ω) can also be represented in the
block form. The operators A ∧ A : C a (Ω × Ω) → C a (Ω × Ω) and a ◦ (A ⊗ A) : C(Ω × Ω) →
C a (Ω × Ω) are defined in the same way.
4. Spectrum of the tensor square of linear operators in the spaces L p (Ω) and C(Ω)
As usual, we will denote by σ(A) the spectrum of the operator A, and by σ p (A) the point
spectrum, that is, the set of all eigenvalues of the operator A. We will denote by σeb (A) the
Browder essential spectrum of the operator A, that is, the set of all points λ ∈ σ(A), such
that at least one of the following conditions holds:
(1) R(A − λI) is not closed;
(2) λis a limit point of σ(A);
(3) n≥0 ker(A − λI)n is of infinite dimension.
Thus σ(A) \ σeb (A) will be the set of all isolated finite-dimensional eigenvalues of the
operator A, (for more detailed information see [6, 7]).
O. Y. Kushel and P. P. Zabreiko 7
In the papers by Ichinose [4–7] there have been obtained the results, representing the
spectra and the parts of the spectra of the tensor product of linear bounded operators
in terms of the spectra and parts of the spectra of the given operators under the natural
suppositions that
(a) the tensor product of linear bounded operators A ⊗ B can be extended from the
set of degenerate functions, and the extension is also a linear bounded operator
in L p (Ω × Ω) and C(Ω × Ω) respectively;
(b) the adjoint spaces L p (Ω × Ω) and rca(Ω × Ω) have the same property.
In fact these statements have been proved in the previous part. The explicit formulae,
expressing the set of all isolated finite dimensional eigenvalues and the Browder essential
spectrum of the operator A ⊗ A in terms of the parts of the spectrum of the given operator
are obtained, for example, in [4, page 110, Theorem 4.2]. In particular, Ichinose proved,
that for the tensor square of a linear bounded operator A the following equalities hold:
σ(A ⊗ A) = σ(A)σ(A);
σ(A ⊗ A) \ σeb (A ⊗ A) = σ(A) \ σeb (A) σ(A) \ σeb (A) \ σeb (A)σ(A) ;
σeb (A ⊗ A) = σeb (A)σ(A).
(4.1)
(4.2)
(4.3)
Besides, for an arbitrary λ ∈ (σ(A ⊗ A) \ σeb (A ⊗ A)) the following equality holds:
ker(A ⊗ A − λI ⊗ I) = ker(A − λi I) ⊗ ker A − λ j I ,
(4.4)
where λi ,λ j ∈ (σ(A) \ σeb (A)) such that λ = λi · λ j .
In the finite-dimensional case, when the matrix A ⊗ A appears to be a tensor square of
the matrix A, Stephanos’s result ([13], see also [10]) tells that the set of all eigenvalues of
the operator A ⊗ A is the set of all the possible products of the form {λi λ j }, where {λi } is
the set of all eigenvalues of the operator A, repeated according to multiplicity. Thus the
property
σ p (A ⊗ A) = σ p (A)σ p (A)
(4.5)
is widely known. In the infinite-dimensional case the analogous formula, expressing
σ p (A ⊗ A) in terms of the parts of the spectrum of the operator A, seems to be unknown.
That is why further we will be interested in the case of a completely continuous operator
A. For a completely continuous operator the following equalities are true:
σ(A) \ σeb (A) \ {0} = σ p (A) \ {0};
σeb (A) = {0} or ∅.
(4.6)
So, from (4.2) we can get the complete information about the nonzero eigenvalues of the
tensor square of a completely continuous operator:
σ p (A ⊗ A) \ {0} = σ p (A)σ p (A) \ {0}.
(4.7)
Here zero can be either a finite- or infinite-dimensional eigenvalue of A ⊗ A, or a point of
the essential spectrum. That is why, even for the case of a completely continuous operator,
formula (4.4) in general is incorrect.
8
Gantmacher-Kreı̆n theorem for 2 nonnegative operators
5. Spectrum of the exterior square of linear operators in the spaces L p (Ω) and C(Ω)
For the exterior square, which is the restriction of the tensor square, the following inclusions are true:
σ(A ∧ A) ⊂ σ(A ⊗ A);
(5.1)
σ p (A ∧ A) ⊂ σ p (A ⊗ A).
(5.2)
In the finite-dimensional case, it is known that the matrix A ∧ A in a suitable basis
appears to be the second-order associated matrix to the matrix A, and we conclude that
all the possible products of the type {λi λ j }, where i < j, form the set of all eigenvalues of
the exterior square A ∧ A, repeated according to multiplicity (see [3, Theorem 23, page
80]).
In the infinite-dimensional case we can also obtain some information concerning
eigenvalues of the exterior square of a linear bounded operator.
Theorem 5.1. Let X be either L p (Ω) or C(Ω) and let {λi } be the set of all eigenvalues of the
operator A : X → X, repeated according to multiplicity. Then all the possible products of the
form {λi λ j }, where i < j, will be the eigenvalues of the exterior square A ∧ A.
Proof. Let λi ,λ j ∈ σ p (A). Then there exist functions x(t), y(t) from X, such that (A −
λi I)x(t) = 0 and (A − λ j I)y(t) = 0. Let us examine the value of the operator A ∧ A −
λi λ j I ∧ I on the degenerate antisymmetric function (x ∧ y)(t1 ,t2 ):
A ∧ A − λi λ j I ∧ I x ∧ y = Ax ∧ Ay − λi λ j x ∧ y
= Ax ∧ Ay − λi x ∧ Ay + λi x ∧ Ay − λi λ j x ∧ y
[because of the linearity of the exterior product]
(5.3)
= Ax − λi x ∧ Ay + λi x ∧ Ay − λ j y = 0.
From this we see that λi λ j ∈ σ p (A ∧ A).
However, just as in the case of the tensor square of an operator, in order to obtain a
statement, analogous to the finite-dimensional case, we need the additional assumption
about complete continuity. For nonzero eigenvalues of the exterior square of a complete
continuous operator, the following statement holds.
Theorem 5.2. Let X be either L p (Ω) or C(Ω) and let {λi } be the set of all eigenvalues of
an absolutely continuous operator A : X → X, repeated according to multiplicity. Then all
the possible products of the type {λi λ j }, where i < j, form the set of all the possible (except,
probably, zero) eigenvalues of the exterior square A ∧ A, repeated according to multiplicity.
Proof. The inclusion {λi λ j }i< j ⊂ σ p (A ∧ A), that is, each product of the form λi λ j , where
i < j, is an eigenvalue of A ∧ A, comes out from Theorem 5.1.
Now we will prove the reverse inclusion: σ p (A ∧ A) ⊂ {λi λ j }i< j . As it was shown above
in formulae (4.7) and (5.2),
σ p (A ∧ A) \ {0} ⊂ σ p (A ⊗ A) \ {0} = σ p (A)σ p (A) \ {0},
(5.4)
O. Y. Kushel and P. P. Zabreiko 9
that is, the operator A ∧ A has no other eigenvalues, except products of the form λi λ j .
Enumerate the set of pairs {(i, j)}i, j = 1,2,.... In this way we get a numeration of
{λi λ j }—the set of all eigenvalues of A ⊗ A, repeated according to multiplicity. Decompose the obtained finite or countable set of indices Λ in the following way:
Λ = Λ1 ∪ Λ2 ∪ Λ3 ,
(5.5)
where the set Λ1 includes the numbers of those pairs (i, j) for which i < j, Λ2 includes
those pairs for which i = j, and Λ3 includes those pairs for which i > j. The set of all
eigenvalues of A ⊗ A, repeated according to multiplicity, will be then decomposed into
three parts:
λα
α ∈Λ
= λ α α ∈Λ 1 ∪ λ α α ∈Λ 2 ∪ λ α α ∈Λ 3 .
(5.6)
As it was shown in Section 3, the operator A ⊗ A has a block structure, and so σ p (A ⊗ A)
can be decomposed into two subsets:
σ p (A ⊗ A) = σ p (A ∧ A) ∪ σ p A ⊗ A|X s ,
(5.7)
where X s is the subset of all symmetric functions from X. In order to prove that the
eigenvalues of A ⊗ A, belonging to the sets {λα }α∈Λ2 and {λα }α∈Λ3 , will not be the eigenvalues of A ∧ A, it is enough to show that they will be the eigenvalues of (A ⊗ A)|X s .
Indeed, let xi (t) ∈ X be an eigenfunction of the operator A, corresponding to the eigenvalue λi . Let us examine the value of the operator (A ⊗ A − λ2i I ∧ I) on the function
xi (t1 )xi (t2 ) ∈ X s (Ω × Ω):
A ⊗ A − λ2i I ⊗ I xi t1 xi t2
= Axi t1 Axi t2 − λ2i xi t1 xi t2
= Axi t1 Axi t2 − λi xi t1 Axi t2 + λi xi t1 Axi t2 − λ2i xi t1 xi t2
= Axi − λi xi t1 Axi t2 + λi xi t1 Axi − λi xi t2 = 0.
(5.8)
From this we see that λ2i ∈ σ p ((A ⊗ A)|X s ). In an analogous way we can prove that a product of the form λi λ j will also be an eigenvalue of (A ⊗ A)|X s (with the corresponding
symmetric function xi (t1 )x j (t2 ) + x j (t1 )xi (t2 )).
It is obvious that the spectrum of the operator a ◦ (A ⊗ A) coincides with the spectrum
of A ∧ A.
6. Generalization of the Gantmacher-Kreı̆n theorems in the case of
2-totally-nonnegative operators in the spaces L p (Ω) and C(Ω)
Let us prove some generalizations of the Gantmacher-Kreı̆n theorems in the case of operators in the spaces L p (Ω) and C(Ω), using the Kreı̆n-Rutman theorem (see, e.g., [14])
about completely continuous operators leaving invariant an almost-reproducing cone K
in a Banach space (for such operators we have the following property of the spectral radius: ρ(A) ∈ σ p (A)).
10
Gantmacher-Kreı̆n theorem for 2 nonnegative operators
Theorem 6.1. Let X be either L p (Ω) or C(Ω), and, respectively, let X be either L p (W)
or C0 (W). Let a completely continuous operator A : X → X with ρ(A) > 0 leave invariant
an almost-reproducing cone K in X and there is only one eigenvalue on the spectral circle
λ = ρ(A). Let its exterior square A ∧ A : X → X leave invariant an almost-reproducing cone
besides ρ(A ∧ A) > 0, and there is also only one eigenvalue on the spectral circle λ =
K in X,
j
ρ(∧ A). Then the operator A has a positive eigenvalue λ1 = ρ(A), and its second (according
to the module) eigenvalue λ2 is also positive.
Proof. Enumerate eigenvalues of a completely continuous operator A, repeated according
to multiplicity, in order of decrease of their modules:
λ 1 < λ 2 < λ 3 ≥ · · · .
(6.1)
Applying the Kreı̆n-Rutman theorem to A, we get λ1 = ρ(A) > 0. Now applying the Kreı̆nRutman theorem to the operator A ∧ A (that obviously is also completely continuous),
we get ρ(A ∧ A) ∈ σ p (A ∧ A).
As it follows from the statement of Theorem 5.2, the exterior square of the operator
A has no other nonzero eigenvalues, except all the possible products of the form λi λ j ,
where i < j. So, we get a conclusion that ρ(A ∧ A) > 0 can be represented in the form
of the product λi λ j with some values of the indices i, j, i < j, and from the fact that
eigenvalues are numbered in a decreasing order, it follows that ρ(A ∧ A) = λ1 λ2 . Therefore
λ2 = ρ(A ∧ A)/λ1 > 0.
To this end a nonnegative linear operator A is called indecomposable (see [2]) if it does
not have any invariant components. For a linear operator which is indecomposable and
nonnegative with respect to the cone of nonnegative functions in L p (Ω) (C(Ω)) and such
that ρ(A) > 0, the positiveness and the simplicity of the first eigenvalue ρ(A) is proved,
for example, in [1, 2, 9, 11, 12]. An indecomposable operator A is called primitive if its
peripheral spectrum consists of the single point ρ(A) and is called imprimitive if its peripheral spectrum contains more than one point. For imprimitive operators that are nonnegative with respect to the cone of nonnegative functions in L p (Ω) (C(Ω)), the invariance of the spectrum of the operator A with respect to some rotation is proved in [1, 8].
(In [1] an analogue of the classical Frobenius theorem on the general form of primitive
and imprimitive matrices is proved for compact indecomposable integral operators. This
statement holds also true for arbitrary, not necessarily integral, compact indecomposable
operators.) Call a cone K in L p (Ω) (C(Ω)) assumed if an indecomposable, nonnegative
with-respect-to-the-cone-K, and completely continuous operator A has the properties,
proved in [1], that is, ρ(A) is a positive simple eigenvalue of A; and if A has h eigenvalues, equal in modulus to ρ(A), then each of them is simple and they coincide with the
hth roots of ρ(A)h . We will call the operator A 2-totally nonnegative if A and A ∧ A are
nonnegative with respect to some assumed cones K and K in L p (Ω) (C(Ω)) and L p (W)
(C0 (W), resp.) and primitive.
Let us prove a generalization of one of the statements of Kreı̆n and Gantmakher in the
case of 2-totally-nonnegative operators in the spaces L p (Ω) and C(Ω).
O. Y. Kushel and P. P. Zabreiko 11
Theorem 6.2. Let X be either L p (Ω) or C(Ω), and, respectively, let X be either L p (W)
or C0 (W). Let a completely continuous operator A : X → X with ρ(A) > 0 be nonnegative
with respect to an assumed cone K in X and indecomposable, and let the operator A ∧ A :
X → X with ρ(A ∧ A) > 0 be nonnegative with respect to an assumed cone K in X and also
indecomposable. Let h(A) and h(A ∧ A) be the indices of imprimitivity of A and A ∧ A
respectively. Then
(a) either h(A) = 1 and h(A ∧ A) is arbitrary, or h(A) = 3 and h(A ∧ A) = 3;
(b) if h(A) = 1 then the operator A has two possible simple eigenvalues λ1 , λ2 , with
ρ(A) = λ1 > λ2 ≥ |λ3 | ≥ · · · ;
(6.2)
(c) if h(A) = h(A ∧ A) = 1, then λ2 is different according to the module from the other
eigenvalues;
(d) if h(A) = 1 and h(A ∧ A) > 1, then the operator A has h(A ∧ A) eigenvalues λ2 ,
λ3 ,...,λh(A∧A)+1 , equal in modulus to λ2 , each of them is simple, and they coincide
with the h(A ∧ A)th roots of λ2h(A∧A) .
Proof. (a) First we will prove that if a completely continuous nonnegative operator A
is imprimitive with h(A) = 2, then its exterior square can not be nonnegative. Really,
according to the theorem of indecomposable operators, we have that there are two eigenvalues ρ(A) > 0 and −ρ(A) on the spectral circle of the operator A. As it follows, there is
only one negative eigenvalue −ρ2 (A) on the spectral circle of A ∧ A and that is impossible
if A ∧ A is nonnegative.
Let h(A) > 3 and let A ∧ A be nonnegative. Prove that A ∧ A is decomposable (i.e., it
has invariant components). Suppose the opposite: let A ∧ A be indecomposable. Then
ρ(A ∧ A) = ρ(A)2 and all the other eigenvalues on the spectral circle of A ∧ A are simple. On the other hand, from Theorem 5.2 and imprimitivity of A, it follows that all
the eigenvalues of A ∧ A, situated on the spectral circle, can be represented as couple
products of different h(A)th roots of ρ(A)h(A) . Let us examine λ j = ρ(A)e2π( j −1)i/h(A) ( j =
1,...,h(A))—all the eigenvalues of A, situated on the spectral circle. It is obvious that
λ2 λh(A) = λ3 λh(A)−1 = · · · = λk λh(A)−(k−2) = · · · = ρ(A)2 . As it follows, ρ(A ∧ A) = ρ(A)2
is not a simple eigenvalue of A ∧ A.
Prove that if A is imprimitive with its index h(A) = 3 and its exterior square is indecomposable, then A ∧ A is also imprimitive with h(A ∧ A) = 3. Indeed, in this case there
are three eigenvalues λ1 = ρ(A), λ2 = ρ(A)e2πi/3 , λ3 = ρ(A)e4πi/3 on the spectral circle of
the operator A and there are also three eigenvalues λ1 λ2 = ρ(A)2 e2πi/3 , λ1 λ3 = ρ(A)2 e4πi/3 ,
and λ2 λ3 = ρ(A)e2πi/3 ρ(A)e4πi/3 = ρ(A)2 that coincide with the 3rd roots of (ρ(A)2 )3 , on
the spectral circle of A ∧ A.
(b) The existence and the positiveness of the first and the second eigenvalues follow
from Theorem 6.1. The simplicity of λ2 follows from the equality λ2 = ρ(A ∧ A)/ρ(A)
and the simplicity of eigenvalues ρ(A) and ρ(A ∧ A).
(c) In the case of h(A) = h(A ∧ A) = 1 the distinction according to the module of λ2
from other eigenvalues is obvious.
12
Gantmacher-Kreı̆n theorem for 2 nonnegative operators
(d) In case of h(A ∧ A) > 1 from Theorem 5.2 and the properties of the peripheral
spectrum of the imprimitive operator A ∧ A it follows that for an eigenvalue λ j , j =
2,...,h(A ∧ A) + 1 the next equality holds: λ j = ρ(A ∧ A)e2π( j −1)i/h(A∧A) /ρ(A).
Consider an example of operator for which the conditions of Theorem 6.2 are satisfied
and h(A) = 3. Let the operator A : R3 → R3 be defined by the matrix
⎛
⎞
0 0 1
⎜
⎟
A = ⎝1 0 0⎠ .
0 1 0
(6.3)
It is obvious that this operator is nonnegative with respect to the cone of nonnegative vectors in R3 and imprimitive with h(A) = 3. In the basis, which consists of exterior products
of the given basic vectors, the matrix of the exterior square of operator A ∧ A coincides
with the second associated matrix, that is, it can be represented in the following form:
⎛
⎞
0 −1 0
⎜
⎟
A ∧ A = ⎝0 0 −1⎠ .
1 0
0
(6.4)
It is obvious that A ∧ A is imprimitive with h(A ∧ A) = 3. It is also obvious that it leaves
invariant the cone of vectors (1,0,0), (0, −1,0), and (0,0,1).
Given an operator A in the finite-dimensional space R3 , it is not difficult, using the
standard scheme, to define a linear integral operator with the same properties, acting in
L p (Ω) or C(Ω).
7. Linear integral operators in the spaces L p (Ω) and C(Ω)
Let us examine a linear integral operator A with kernel k(t,s), acting in the space L p (Ω).
Observing that
(A ⊗ A)x t1 ,t2 = A(1) A(2) x t1 ,t2
=
=
k t1 ,s1
Ω Ω Ω
Ω
k t2 ,s2 x s1 ,s2 dμ dμ
(7.1)
k t1 ,s1 k t2 ,s2 x s1 ,s2 d(μ ⊗ μ) s1 ,s2 ,
we conclude that the tensor square of the operator A is a linear integral operator with
kernel k(t1 ,s1 )k(t2 ,s2 ), acting in the space L p (Ω × Ω). Thus the exterior square of A
is a restriction of the integral operator with kernel k(t1 ,s1 )k(t2 ,s2 ) onto the subspace
Lap (Ω × Ω).
O. Y. Kushel and P. P. Zabreiko 13
Let us examine the value of the operator a ◦ (A ⊗ A) on an arbitrary function x(t1 ,t2 )
from L p (Ω × Ω):
a◦
Ω×Ω
=
1
2
Ω×Ω
1
2
−
1
=
2
k t1 ,s1 k t2 ,s2 x s1 ,s2 d(μ ⊗ μ) s1 ,s2
k t1 ,s1 k t2 ,s2 x s1 ,s2 d(μ ⊗ μ) s1 ,s2
Ω×Ω
k t2 ,s1 k t1 ,s2 x s1 ,s2 d μ ⊗ μ s1 ,s2
k t ,s 1 1
Ω×Ω k t2 ,s1
(7.2)
k t1 ,s2 x s1 ,s2 d(μ ⊗ μ) s1 ,s2 .
k t2 ,s2 Since the values of the operators a ◦ (A ⊗ A) and A ∧ A on antisymmetric functions do
coincide, it is obvious that
Ω×Ω
k t1 ,s1 k t2 ,s2 x s1 ,s2 d(μ ⊗ μ) s1 ,s2
= a◦
Ω×Ω
k t1 ,s1 k t2 ,s2 x s1 ,s2 d(μ ⊗ μ) s1 ,s2
k t ,s 1 1
1
=
2 Ω×Ω k t2 ,s1
k t1 ,s2 x s1 ,s2 d(μ ⊗ μ) s1 ,s2 ,
k t2 ,s2 (7.3)
where x(t1 ,t2 ) is an arbitrary function from Lap (Ω × Ω).
k(t1 ,s1 ) k(t1 ,s2 )
the second associated kernel k(t,s) and
Further, we will call the kernel k(t
2 ,s1 ) k(t2 ,s2 )
will denote it by (k ∧ k)(t1 ,t2 ,s1 ,s2 ).
we have that
Since Ω × Ω = W ∪ W,
(A ∧ A)x t1 ,t2
=
1
2
W
1
+
2
(k ∧ k) t1 ,t2 ,s1 ,s2 x s1 ,s2 d(μ ⊗ μ) s1 ,s2
W
(k ∧ k) t1 ,t2 ,s1 ,s2 x s1 ,s2 d(μ ⊗ μ) s1 ,s2
(7.4)
= ··· .
Exchanging the places of s1 and s2 in the second integral, we get
··· =
W
(k ∧ k) t1 ,t2 ,s1 ,s2 x s1 ,s2 d(μ ⊗ μ) s1 ,s2 .
(7.5)
Now we can consider the exterior square of the operator A, acting in the space L p (Ω), as
a linear integral operator, acting in L p (W), with the kernel equal to the second associated
to k(t,s). The same reasoning is true for the space C(Ω).
A nonnegative kernel k(t,s) is called indecomposable if for any measurable set D ∈
Ω, 0 < μ(D) < μ(Ω), there exist measurable sets A ∈ D, B ∈ Ω \ D, such that μ(A) > 0,
14
Gantmacher-Kreı̆n theorem for 2 nonnegative operators
μ(B) > 0 and k(t,s) > 0 almost everywhere on A × B. A nonnegative integral operator A
with a kernel k(t,s) is indecomposable if and only if the kernel k(t,s) is indecomposable
(see [8]). In [1], for an indecomposable nonnegative linear integral operator, the positiveness and the simplicity of the eigenvalue ρ(A) is proved (see [1, page 6, Theorem
2]).
An indecomposable kernel k(t,s) is called imprimitive, if there exists a decomposition
of the set Ω in n > 1 nonintersecting sets Ω j ( j = 1,...,n) of a positive measure: Ω =
∪nj=1 Ω j , for which k(t,s) = 0 with t ∈ Ω j , s ∈
/ Ω j+1 for any j = 1,...,n (in the case of
j = n we presuppose j + 1 to be equal to 1). Otherwise the kernel k(t,s) is called primitive.
A compact integral operator A is imprimitive (resp., primitive) if and only if its kernel
is imprimitive (primitive). A kernel k(t,s) is called 2-totally nonnegative, if both k(t,s)
and (k ∧ k)(t1 ,t2 ,s1 ,s2 ) are primitive. Further, we will examine on the set W the second
associated kernel and the conditions implied to it. The index of imprimitivity of (k ∧
k)(t1 ,t2 ,s1 ,s2 ) will be denoted by h(k ∧ k).
Let the kernel k(t,s) of a completely continuous integral operator A in the space L p (Ω)
or C(Ω) be nonnegative and primitive. Let its second associated kernel be also nonnegative and primitive. Then Theorem 6.2 implies that the second, according to the module,
eigenvalue of the operator A λ2 is positive, simple, and different in modulus from the other
eigenvalues.
Note that in this reasoning the kernel is not presupposed to be continuous, but only
assume that the operator A acts in the space L p (Ω) or C(Ω).
8. Concluding remarks
All the results given in the present paper can be easily spread on k-totally-nonnegative
operators (k > 2), acting in the space L p (Ω) or C(Ω). Similar results will be also true for
other spaces, for example, for some ideal spaces, the Lorenz-Martzinkevich space, and so
forth.
References
[1] E. A. Alekhno and P. P. Zabreiko, Quasipositive elements and indecomposable operators in
ideal spaces. II, Vestsı̄ Natsyyanal’naı̆ Akadèmı̄ı̄ Navuk Belarusı̄. Seryya Fı̄zı̄ka-Matèmatychnykh
Navuk (2003), no. 1, 5–10, 124.
[2] T. Andô, Positive linear operators in semi-ordered linear spaces, Journal of the Faculty of Science,
Hokkaido University. Series I. 13 (1957), 214–228.
[3] F. R. Gantmacher and M. G. Kreı̆n, Oszillationsmatrizen, Oszillationskerne und kleine Schwingungen mechanischer Systeme, Wissenschaftliche Bearbeitung der deutschen Ausgabe: Alfred Stöhr.
Mathematische Lehrbücher und Monographien, I. Abteilung, vol. 5, Akademie, Berlin, 1960.
[4] T. Ichinose, Operators on tensor products of Banach spaces, Transactions of the American Mathematical Society 170 (1972), 197–219.
, Operational calculus for tensor products of linear operators in Banach spaces, Hokkaido
[5]
Mathematical Journal 4 (1975), no. 2, 306–334.
, Spectral properties of tensor products of linear operators. I, Transactions of the American
[6]
Mathematical Society 235 (1978), 75–113.
O. Y. Kushel and P. P. Zabreiko 15
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
, Spectral properties of tensor products of linear operators. II. The approximate point spectrum and Kato essential spectrum, Transactions of the American Mathematical Society 237
(1978), 223–254.
M. A. Krasnosel’skij, Je. A. Lifshits, and A. V. Sobolev, Positive Linear Systems. The Method of
Positive Operators, Sigma Series in Applied Mathematics, vol. 5, Heldermann, Berlin, 1989.
H. P. Lotz, Über das Spektrum positiver Operatoren, Mathematische Zeitschrift 108 (1968), no. 1,
15–32.
T.-W. Ma, Classical Analysis on Normed Spaces, World Scientific, New Jersey, 1995.
F. Niiro and I. Sawashima, On the spectral properties of positive irreducible operators in an arbitrary Banach lattice and problems of H. H. Schaefer, Scientific Papers of the College of Arts and
Sciences. The University of Tokyo 16 (1966), 145–183.
I. Sawashima, On spectral properties of some positive operators, Natural Science Report of the
Ochanomizu University 15 (1964), 53–64.
C. Stéphanos, Sur une extension du calculus des substitutions linéaires, Journal de Mathematiques
Pures et Appliquees (5) 6 (1900), 73–128.
P. P. Zabreiko and S. V. Smitskikh, A theorem of M. G. Krein and M. A. Rutman, Functional
Analysis and Its Applications 13 (1980), 222–223.
O. Y. Kushel: Mechanics and Mathematics Faculty, Belarusian State University, Pr. Independence 4,
220050 Minsk, Belarus
E-mail address: [email protected]
P. P. Zabreiko: Mechanics and Mathematics Faculty, Belarusian State University, Pr. Independence 4,
220050 Minsk, Belarus
E-mail address: [email protected]